“…Therefore it is not considered to be a better choice than PCA. Alternatively to the quantitation approach, an optimal region selection algorithm could be used, 13 which finds the optimum window width rather than a single common width. Figure 6 provides a good indication of why the introduction of the MRI features improves classification.…”
Section: Discussionmentioning
confidence: 99%
“…Several methods to perform data reduction and feature selection have been described in the literature. 6,[11][12][13] The last step in automated tumor classification, involves the actual classifier. In our case the classifier was based on the Mahalanobis distance.…”
The purpose of this paper is to evaluate the effect of the combination of magnetic resonance spectroscopic imaging (MRSI) data and magnetic resonance imaging (MRI) data on the classification result of four brain tumor classes. Suppressed and unsuppressed short echo time MRSI and MRI were performed on 24 patients with a brain tumor and four volunteers. Four different feature reduction procedures were applied to the MRSI data: simple quantitation, principal component analysis, independent component analysis and LCModel. Water intensities were calculated from the unsuppressed MRSI data. Features were extracted from the MR images which were acquired with four different contrasts to comply with the spatial resolution of the MRSI. Evaluation was performed by investigating different combinations of the MRSI features, the MRI features and the water intensities. For each data set, the isolation in feature space of the tumor classes, healthy brain tissue and cerebrospinal fluid was calculated and visualized. A test set was used to calculate classification results for each data set. Finally, the effect of the selected feature reduction procedures on the MRSI data was investigated to ascertain whether it was more important than the addition of MRI information. Conclusions are that the combination of features from MRSI data and MRI data improves the classification result considerably when compared with features obtained from MRSI data alone. This effect is larger than the effect of specific feature reduction procedures on the MRSI data. The addition of water intensities to the data set also increases the classification result, although not significantly. We show that the combination of data from different MR investigations can be very important for brain tumor classification, particularly if a large number of tumors are to be classified simultaneously.
“…Therefore it is not considered to be a better choice than PCA. Alternatively to the quantitation approach, an optimal region selection algorithm could be used, 13 which finds the optimum window width rather than a single common width. Figure 6 provides a good indication of why the introduction of the MRI features improves classification.…”
Section: Discussionmentioning
confidence: 99%
“…Several methods to perform data reduction and feature selection have been described in the literature. 6,[11][12][13] The last step in automated tumor classification, involves the actual classifier. In our case the classifier was based on the Mahalanobis distance.…”
The purpose of this paper is to evaluate the effect of the combination of magnetic resonance spectroscopic imaging (MRSI) data and magnetic resonance imaging (MRI) data on the classification result of four brain tumor classes. Suppressed and unsuppressed short echo time MRSI and MRI were performed on 24 patients with a brain tumor and four volunteers. Four different feature reduction procedures were applied to the MRSI data: simple quantitation, principal component analysis, independent component analysis and LCModel. Water intensities were calculated from the unsuppressed MRSI data. Features were extracted from the MR images which were acquired with four different contrasts to comply with the spatial resolution of the MRSI. Evaluation was performed by investigating different combinations of the MRSI features, the MRI features and the water intensities. For each data set, the isolation in feature space of the tumor classes, healthy brain tissue and cerebrospinal fluid was calculated and visualized. A test set was used to calculate classification results for each data set. Finally, the effect of the selected feature reduction procedures on the MRSI data was investigated to ascertain whether it was more important than the addition of MRI information. Conclusions are that the combination of features from MRSI data and MRI data improves the classification result considerably when compared with features obtained from MRSI data alone. This effect is larger than the effect of specific feature reduction procedures on the MRSI data. The addition of water intensities to the data set also increases the classification result, although not significantly. We show that the combination of data from different MR investigations can be very important for brain tumor classification, particularly if a large number of tumors are to be classified simultaneously.
“…However, this is limited by the integrity of the assignments, and potentially valuable information is disregarded if the spectra contain metabolite signals that are not accounted for in the basis set. A comparison of these two different approaches to classification using MRS was beyond the scope of this study; however, future studies should investigate alternative spectral featureextraction methods that do not involve fitting to estimate metabolite concentrations (14,36).…”
1 H MRS has great potential for the clinical investigation of childhood brain tumours, but the low incidence in, and difficulties of performing trials on, children have hampered progress in this area. Most studies have used a long-TE, thus limiting the metabolite information obtained, and multivariate analysis has been largely unexplored. Thirty-five children with untreated cerebellar tumours (18 medulloblastomas, 12 pilocytic astrocytomas and five ependymomas) were investigated using a single-voxel short-TE PRESS sequence on a 1.5 T scanner. Spectra were analysed using LCModel TM to yield metabolite profiles, and key metabolite assignments were verified by comparison with high-resolution magic-angle-spinning NMR of representative tumour biopsy samples. In addition to univariate metabolite comparisons, the use of multivariate classifiers was investigated. Principal component analysis was used for dimension reduction, and linear discriminant analysis was used for variable selection and classification. A bootstrap cross-validation method suitable for estimating the true performance of classifiers in small datasets was used. The discriminant function coefficients were stable and showed that medulloblastomas were characterised by high taurine, phosphocholine and glutamate and low glutamine, astrocytomas were distinguished by low creatine and high N-acetylaspartate, and ependymomas were differentiated by high myo-inositol and glycerophosphocholine. The same metabolite features were seen in NMR spectra of ex vivo samples. Successful classification was achieved for glial-cell (astrocytoma ĂŸ ependymoma) versus non-glial-cell (medulloblastoma) tumours, with a bootstrap 0.632 ĂŸ error, e B.632ĂŸ , of 5.3%. For astrocytoma vs medulloblastoma and astrocytoma vs medulloblastoma vs ependymoma classification, the e B.632ĂŸ was 6.9% and 7.1%, respectively. The study showed that 1 H MRS detects key differences in the metabolite profiles for the main types of childhood cerebellar tumours and that discriminant analysis of metabolite profiles is a promising tool for classification. The findings warrant confirmation by larger multi-centre studies.
“…The examples of successful methods to find discriminative spectral regions are an Optimal Region Selector (ORS) [1] guided by a genetic algorithm, a top-down and bottom-up multiresolution feature extraction algorithms proposed by Kumar et al [2], Recursive Band Selection (RBE) [3] etc. The advantage of these techniques is that they make use of the connectivity between neighbouring spectral bins when finding discriminative groups of spectral bands, while the standard feature reduction approaches (such as forward/backward feature selection or PCA [4]) neglect the apriori available information on the ordering of spectral wavelengths.…”
Abstract. In the past few years a variety of successful algorithms to select/extract discriminative spectral bands was introduced. By exploiting the connectivity of neighbouring spectral bins, these techniques may be more beneficial than the standard feature selection/extraction methods applied for spectral classification. The goal of this paper is to study the effect of the training sample size on the performance of different strategies to select/extract informative spectral regions. We also consider the success of these methods compared to Principal Component Analysis (PCA) for different numbers of extracted components/groups of spectral bands.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citationsâcitations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.